ARCTIC
VOL. 58, NO. 2 (JUNE 2005) P. 203 – 214
Demography and Viability of a Hunted Population of Polar Bears
MITCHELL K. TAYLOR,1 JEFF LAAKE,2 PHILIP D. McLOUGHLIN,3,4 ERIK.W. BORN,5 H. DEAN CLUFF,6
STEVEN H. FERGUSON,3,7 AQQALU ROSING-ASVID,5 RAY SCHWEINSBURG6,8 and FRANÇOIS MESSIER3
(Received 26 July 2004; accepted in revised form 15 December 2004)
ABSTRACT. We estimated demographic parameters and harvest risks for a population of polar bears (Ursus maritimus)
inhabiting Baffin Bay, Canada and Greenland, from 1974 to 1997. Our demographic analysis included a detailed assessment of
age- and sex-specific survival and recruitment from 1221 marked polar bears, which used information contained within the
standing age distribution of captures and mark-recapture analysis performed with Program MARK. Unharvested (natural)
survival rates for females (± 1 SE) from mark-recapture analysis were 0.620 ± 0.095 (cubs), 0.938 ± 0.042 (ages 1 – 4), 0.953 ±
0.020 (ages 5 – 20), and 0.919 ± 0.046 (ages 21+). Total (harvested) survival rates for females were reduced to 0.600 ± 0.096 (cubs),
0.901 ± 0.045 (ages 1 – 4), 0.940 ± 0.021 (ages 5–20), and 0.913 ± 0.047 (ages 21+). Mean litter size was 1.59 ± 0.07 cubs, with
a mean reproductive interval of 2.5 ± 0.01 years. By age 5, on average 0.88 ± 0.40 of females were producing litters. We estimated
the geometric means (± bootstrapped SDs) for population growth rates at stable age distribution as 1.055 ± 0.011 (unharvested)
and 1.019 ± 0.015 (harvested). The model-averaged, mark-recapture estimate of mean abundance (± 1 SE) for years 1994 – 97 was
2074 ± 266 bears, which included 1017 ± 192 females and 1057 ± 124 males. We incorporated demographic parameters and their
error terms into a harvest risk analysis designed to consider demographic, process, and sampling uncertainty in generating
likelihoods of persistence (i.e., a stochastic, harvest-explicit population viability analysis). Using our estimated harvest of polar
bears in Baffin Bay (88 bears/yr), the probability that the population would decline no more than could be recovered in five years
was 0.95, suggesting that the current hunt is sustainable.
Key words: demography, harvest, mark-recapture, polar bear, Ursus maritimus, population viability analysis, program MARK,
recruitment, survival
RÉSUMÉ. De 1974 à 1997, on a évalué les paramètres démographiques d’une population d’ours polaires (Ursus maritimus)
habitant la baie de Baffin (Canada et Groenland), ainsi que les risques associés à leur prélèvement. Notre analyse démographique
comprenait un bilan détaillé de la survie et du recrutement par âge et par sexe, bilan mené sur 1221 ours polaires étiquetés et qui
faisait appel à l’information contenue dans les limites de la structure d’âge des captures à un moment précis, ainsi que des analyses
de marquage-recapture réalisées avec le logiciel MARK. Les taux de survie sans prélèvements (c’est-à-dire naturels) des femelles
(± 1 erreur-type) tirés de l’analyse de marquage-recapture étaient les suivants: 0,620 ± 0,095 (oursons), 0,938 ± 0,042 (1 – 4 ans),
0,953 ± 0,020 (5–20 ans) et 0,919 ± 0,046 (21 ans et plus). Les taux de survie globaux (avec prélèvements) des femelles diminuaient
à: 0,600 ± 0,096 (oursons), 0,901 ± 0,045 (1 – 4 ans), 0,940 ± 0,021 (5–20 ans) et 0,913 ± 0,047 (21 ans et plus). La taille moyenne
des portées était de 1,59 ± 0,07 ourson avec des intervalles moyens de reproduction de 2,5 ± 0,01 ans. Arrivées à l’âge de cinq
ans, en moyenne 0,88 ± 0,40 des femelles avaient eu des petits. On a évalué que les moyennes géométriques (± écart-type
bootstrappé) pour les taux de croissance de la population à la structure d’âge stable étaient de 1,055 ± 0,011 (sans prélèvements)
et de 1,019 ± 0,015 (avec prélèvements). La valeur estimée à partir du marquage-recapture, moyennée par le modèle, de
l’abondance moyenne (± 1 erreur-type), pour les années allant de 1994 à 1997 était de 2074 ± 266 ours, dont 1017 ± 192 femelles
et 1057 ± 124 mâles. On a intégré les paramètres démographiques et leurs termes d’erreur dans une analyse des risques de
prélèvements conçue pour tenir compte des incertitudes démographiques, de processus et d’échantillonnage lors du calcul des
probabilités de persistance (c.-à-d. une analyse stochastique de la viabilité de la population qui tient compte des prélèvements).
En se basant sur nos prélèvements estimés de l’ours polaire dans la baie de Baffin (88 ours/an), la probabilité que la population
ne décline pas plus que ce qu’elle pourrait récupérer en 5 ans était de 0,95, ce qui suggère que la chasse actuelle est durable.
1
Department of the Environment, Government of Nunavut, P.O. Box 209, Igloolik, Nunavut X0A 0L0, Canada
National Marine Mammal Laboratory, Alaska Fisheries Science Center, National Marine Fisheries Service, Seattle, Washington 98115,
U.S.A.
3
Department of Biology, University of Saskatchewan, 112 Science Place, Saskatoon, Saskatchewan S7N 5E2, Canada
4
Corresponding author: mcloughlin@sask.usask.ca
5
Greenland Institute of Natural Resources, P.O. Box 570, DK-3900 Nuuk, Greenland
6
Department of Resources, Wildlife, and Economic Development, Government of the Northwest Territories, Yellowknife, Northwest
Territories X1A 2P9, Canada
7
Current address: Fisheries and Oceans Canada, 501 University Crescent, Winnipeg, Manitoba R3T 2N6, Canada
8
Current address: Arizona Game and Fish Department, WMRS 2221, West Greenway Road, Phoenix, Arizona 85023, U.S.A.
© The Arctic Institute of North America
2
204 • M.K. TAYLOR et al.
Mots clés: démographie, prélèvement, marquage-recapture, ours polaire, viabilité de la population, analyse, logiciel MARK,
recrutement, survie, Ursus maritimus
Traduit pour la revue Arctic par Nésida Loyer.
INTRODUCTION
Stabilizing populations through sustained harvest is a
basic application of wildlife management theory; in practice, however, it is exceedingly difficult to establish the
correct composition (sex and age distribution) of sustainable yields. Traditional deterministic population models
require considerable information to identify sustainable
yields (Caughley, 1977), including the natural birth and
death rates for different age/sex strata in the population,
the shape of the relationship between population growth
rate in the absence of harvest and population size, or
parameters such as carrying capacity or intrinsic rate of
increase. It is particularly important, however, that this
information be provided with little or no error. If biologists
are uncertain about input parameters, deterministic models of yield will fail to provide results applicable to the
management of real populations. Since vital rates of actual
populations are almost always obtained with some degree
of error, either through sampling error or through observations of process variation (e.g., inter-year variation due to
environmental stochasticity; White, 2000), deterministic
attempts to calculate sustainable yield are only rarely of
value for managing real populations.
Rather than relying upon deterministic models to establish sustainable yields, an alternative approach may be to
manage for harvests that will provide a reasonable probability of population persistence for some time into the
future. Models of probability of population persistence,
such as stochastic population viability analysis (PVA), are
ideal for incorporating parameter variation into harvest
models, including demographic, environmental, and sampling variation (review in White, 2000). When information is uncertain, theoretically any harvest level poses
some risk to a population. PVA has the flexibility to
provide managers and stakeholders with a probability that
a given harvest composition will be sustainable, but also
with consequences (e.g., the length of moratorium required to restore the population) should a harvest later be
deemed too severe (Taylor et al., 2002).
We used mark-recapture data collected from 1974 to
1997 to estimate the demographic characteristics of polar
bears inhabiting Baffin Bay, Canada and Greenland
(Fig. 1), and harvest risks for the population at its estimated size. Our demographic analysis includes a detailed
assessment of age- and sex-specific survival and recruitment from 1221 marked polar bears, which used information contained within the standing age distribution of
captures and survival and abundance estimates from markrecapture analysis. We incorporated demographic parameters and their error terms into a harvest risk analysis
FIG. 1. The location of all captures and recaptures within the Baffin Bay polar
bear population, northern Canada and Greenland. Boundaries are defined as in
Taylor et al. (2001a). Spring captures from 1974 to 1993 are indicated by the
symbol (). Autumn (onshore) captures from 1993 to 1997 are indicated by the
symbol ().
designed to consider demographic, environmental, and
sampling variation in generating likelihoods of persistence (i.e., a stochastic, harvest-explicit PVA). We provide
estimates of probability of population persistence above a
threshold of population decline for an array of harvest
rates, but also the estimated numbers of years required for
recovery of the population should a harvest strategy result
in population decline below the stated threshold. We
contrast the mark-recapture analysis presented in this
study with an earlier analysis of the demography of polar
bears in Baffin Bay to illustrate the value of recent advances in mark-recapture methodology.
METHODS
Study Area
The geographic bounds of the Baffin Bay polar bear
population (Fig. 1) have been previously evaluated using
movements of marked and recaptured (or harvested) individuals (Taylor and Lee, 1995), DNA analysis (Paetkau et
al., 1999), and movements of radio-collared adult females
(Taylor et al., 2001a). The Baffin Bay population is considered to be demographically discrete, but shared between Greenland and Canada. The total geographic area
of the Baffin Bay population is approximately 1.03 ×
106 km2.
POLAR BEAR DEMOGRAPHY AND POPULATION VIABILITY • 205
The distribution of polar bears in Baffin Bay is strongly
influenced by the seasonal distribution of sea ice (Ferguson
et al., 2000; Taylor et al., 2001a). Because of currents and
prevailing winds in this part of the Arctic, sea ice remains
longest each spring along the Baffin Island coast. As a
result, almost every polar bear of the Baffin Bay population concentrates in spring near Baffin Island, being forced
ashore for the open-water season (Taylor et al., 2001a).
This feature of the Baffin Bay population greatly enhanced
our efforts at uniform sampling of the population. Nondenning bears (i.e., males and non-parturient females)
return to the sea ice in November (Ferguson et al., 1997),
and many proceed across Baffin Bay to Greenland waters
(but return west in spring, as sea ice disappears first from
east Baffin Bay and Davis Strait). Most Baffin Bay polar
bears do not move south except along the Baffin Island
coast because of the open-water barrier caused by the West
Greenland Current.
Captures, Recaptures, and Recoveries
Four main capture programs in Baffin Bay yielded the
data used in this study. The first effort (1974 – 79) was
incidental to a study of the Lancaster Sound population
and occurred only from communities on northern Baffin
Island. From 1980 to 1985, polar bears were captured as
part of an independent demographic study on east-central
Baffin Island. From 1989 to 1993, there were incidental
captures in both Greenland and Canada associated with
movement (radio-telemetry) studies. The most recent capture program was staged from 1994 to 1997, but with little
effort in 1996. During 1974 – 93, capture sampling was
conducted in spring (April–May) prior to breakup by
searching shorefast ice along Baffin Island. Active pack
ice was not searched because of difficulties in tracking and
fuel limitations of search helicopters. During 1993 – 97,
capture sampling was conducted in autumn (September–
October), during the open-water season, when all Baffin
Bay polar bears were onshore. (Sampling took place during both seasons in 1993.) The coastal islands, coastal
areas, and the interior of Baffin Island were systematically
and uniformly searched in 1994, 1995, and 1997. All
captures were opportunistic: every bear seen was captured
(or recaptured).
We chemically immobilized all bears and their dependent cubs for capture and marking according to procedures
described by Stirling et al. (1989), following Animal Care
Protocol No. 950005 of the University of Saskatchewan
and under guidance of the Canadian Council on Animal
Care. Bears captured or recaptured prior to 1985 were
immobilized primarily with Sernylan (Furnell and
Schweinsburg, 1984); bears captured in later years were
immobilized with Telazol (Stirling et al., 1989). We assigned a unique identification number to each bear upon its
initial capture and marked the animals accordingly, using
plastic ear tags and permanent lip tattoos. We also marked
each bear with a wax crayon on the fur to ensure that it was
not captured more than once per year. We considered a
bear’s age as “known” if the bear was captured as a cub-ofthe-year (cub) or yearling, or if its age was estimated by
counting annular rings of an extracted vestigial premolar
(Calvert and Ramsay, 1998). We recorded the sex, age,
family status, and location of all polar bears killed by
hunters, killed as problem bears, or found dead from any
cause. We believe Greenland hunters reported most of
their kill of marked bears (E.W. Born, pers. obs.).
Reproduction
Our estimates of reproductive parameters were developed from analysis of the standing age distribution of the
population, and they focused on the Baffin Bay population
structure during the largest and most recent capture program (1994 – 97). We determined the standing age distribution for the population based only on captures and
recaptures that occurred inside the Baffin Bay boundary
(Fig. 1). To acknowledge parental care, we based agespecific litter production rate on the number of females
that had been available to mate the previous year (i.e.,
those females with no cubs or two-year-olds), rather than
the total number of females. Our estimate of litter production rate assumed that females survived after mating. We
included litter production rate for age 4 and age 5 to
acknowledge potential subadult reproduction. We assumed
that by age 6, all females were producing litters at adult
rates. Our methods for estimating litter production rates
and litter size of cubs and yearlings are described in detail
in Taylor et al. (1987a, 2000). The proportion of cubs (age
0) that were male was the binomial mean M:F ratio based
on the total number of age 0 cubs between 1994 and 1997.
We calculated an additional four summary reproductive
parameters for the Baffin Bay population: mean reproductive (litter production) interval, mean annual reproductive
rate, proportion of females that reached adulthood but had
only one litter, and proportion of females that reached
adulthood but produced no litters.
We used a jackknife method (Arveson, 1969) to obtain
variances of reproductive parameters estimated from the
standing age distribution (Taylor et al., 1987a, 2000). The
data were collected over many years, so our estimates of
variance included both sampling error and inter-year ecological variation. The variances of summary reproductive
parameters were determined by means of Monte Carlo
simulations with 1200 iterations, using the Visual Basic
program “Vital Rates” (Taylor et al., 2000). The program
is available from the senior author upon request.
Survival and Abundance
Estimates of survival and abundance were constructed
from analysis of capture-recapture and recovery data
using the Burnham Cormack-Jolly-Seber (CJS) model
formulation implemented in Program MARK (White et
al., 2000). The Burnham model (Burnham, 1993) incorpo-
206 • M.K. TAYLOR et al.
rates harvest recoveries of tagged animals into a CormackJolly-Seber model framework for capture-recapture data.
The CJS likelihood for capture-recapture data is conditioned on initial capture events (i.e., the initial capture is
treated as a release). The likelihood is based solely on
recapture events of marked (i.e., previously caught) animals, and it is defined by user-specified models for survival (S) and (re)capture probabilities (p) that may be
expressed as functions of covariates such as sex, age, and
time. The Burnham model extends the CJS model to
include a model for recovery (i.e., harvest) probabilities
(r) that may also include covariates.
We used Program MARK to analyze the capture-recapture and recovery data collected in Baffin Bay from 1974
to 1997 (through 1998 for recoveries). Captures of bears
from 1974 to 1992 were used as initial captures, but
recaptures were ignored during this period because captures were generally restricted to a local subset of the
entire population. We used recaptures only from 1993 to
1997, when extensive autumn capturing was conducted.
Recoveries of harvested bears were assumed to occur
between captures; however, since our capture efforts typically spanned a period of two months, there may have been
some overlap between harvesting and capturing. We treated
spring harvest as if it occurred before spring captures and
after autumn captures of the previous year. Eighty-eight
percent of bear harvests occurred between October and
April, and 95% of captures were in April–May and September–October. Thus, harvest recoveries for year i + 1
included autumn harvest recoveries from year i and spring
harvest recoveries from year i + 1. We fixed the fidelity
parameter (F) at 1.0; that is, we assumed no permanent
emigration from Baffin Bay and complete overlap of the
population sampled by capture and harvesting, which was
likely with autumn captures. We used all recaptures and
harvest recoveries of Baffin Bay bears, even if they occurred
outside the boundaries of the Baffin Bay population.
We identified a priori a limited number of parameter
sets (models) consistent with our sampling effort. For the
pooled 1974 – 97 data, we examined a series of models for
capture probability (p) that incorporated potential
covariates for our system. We expected that capture probability would vary by year because conditions were different each year, although the amount of capture effort was
approximately the same for 1994, 1995, and 1997. Therefore, we considered a model that allowed capture probability to vary for each year (pyear) and a null model in which
capture probability did not vary between years. Bears were
located by visual observation and tracking from a helicopter, and successful location and eventual capture were
likely affected by the number of bears in a group, their
reaction (if any) to the capture helicopter, and their movement patterns and fidelity to known high-use areas. Because these factors were likely to vary for different classes
of family status, we categorized bears into three classes—
females, cubs, and yearlings; subadults (2 – 4 years) of
either sex; and adult males—and considered models in
which capture probability varied for these classes (pfamily).
We also considered models (pradio) in which bears with
transmitters (radios) possessed a different and most likely
higher capture probability because their location was known
at various times throughout the year. Cubs and yearlings of
a female with a transmitter were considered to have the
same probability of capture as their mother.
For survival probability (S), we considered models that
included sex, age, and season of mark deployment (spring
or autumn) as covariates (Ssex, Sage, and Sseason). Males were
more likely to be harvested, so we considered sex as a
potential covariate of survival. Survival was also likely to
vary by age; in particular, we expected cubs to have lower
survival than non-cubs. We structured age into four classes:
1) cubs, 2) yearlings/sub-adults (1 – 4 years), 3) adults (5 –
20 years), and 4) senescent adults (21+ years). Although
annual differences in harvest and environmental conditions could potentially create annual variation in survival,
we did not consider models with annual survival rates
because we could not reliably estimate separate rates for
1974 – 92, when we did not recapture bears and there were
too few recoveries. We expected that survival might have
been lower for bears marked in spring (1974 – 92) than for
those marked in autumn, because bears that used shorefast
ice during spring would be vulnerable to harvest in both
spring and autumn, whereas those marked in the autumn
would include bears that did not use shorefast ice in spring
(and would thus be less vulnerable to harvest).
We used the same sub-model for survival (S) and recovery probabilities (r), on the assumptions that harvest was
a primary source of mortality and natural mortality was
relatively constant. In the MARK implementation of the
Burnham CJS model, r is the probability that a dead
marked bear is recovered and reported. We have included
only recoveries of harvested bears, so if all harvests of
marked bears were reported, r represents the proportion of
mortality due to harvest. If all harvested marked bears
were reported, then the probability that a bear would be
harvested was (1 – S)r.
We fitted a series of models using each capture probability sub-model with each survival/recovery probability
sub-model. We considered additive models with main
effects (e.g., S/rsex + age + spring, pyear + family + radio), but we
limited our analysis to relatively simple interactions (e.g.,
S/rsex × age) because data were insufficient to support more
complex models. For model selection, we used Akaike’s
Information Criterion adjusted for overdispersion (QAICc,
Burnham and Anderson, 2002). The data were likely to be
overdispersed (i.e., greater than binomial variation) because survival and capture events of family groups (i.e.,
females with cubs or yearlings) were not independent.
Various approaches for estimating the overdispersion coefficient ( ĉ ) have been suggested (Lebreton et al., 1992;
White et al., 2000). Overdispersion can occur if fates
(survival, capture) are linked. McCullagh and Nelder
(1991:125) showed that the overdispersion coefficient is
bounded above by the cluster (group) size for a population
POLAR BEAR DEMOGRAPHY AND POPULATION VIABILITY • 207
in equal-sized clusters in which the binomial response
varies by cluster. However, this approach could overestimate ĉ if group composition changes over time so that
fates are not linked. Mother-cub pairs are a clear example
of linked fates: cub survival is highly dependent on the
mother’s survival, and a cub will only be recaptured as a
yearling with its mother. While the survival of a yearling
may depend somewhat on its mother’s survival, yearlings
were not likely to be recaptured with their mothers as twoyear-olds. Therefore, we estimated ĉ as the ratio of total
bears captured during 1993 – 97 to non-cubs captured during the same period. This method effectively treated the
sample size as the number of non-cub captures. We used ĉ
to compute QAICc and to inflate variance estimates.
Because the Burnham-CJS model likelihood does not
include the probability distribution for unmarked animals, it
may not be optimal (i.e., of highest precision) for estimating
abundance. However, at present we are unaware of any model
that has been developed to include recoveries in a Jolly-Seber
model formulation with a probability distribution for first
captures. We followed the approach of McDonald and Amstrup
(2001) and Taylor et al. (2002) in constructing abundance
estimates. We used estimated recapture probabilities (pi) in
year i from animals marked prior to year i with the number of
captures and recaptures in year i to estimate abundance (Ni),
for each year. We computed variance estimates for Ni using
a Taylor series approximation (Seber, 1982) (also known as
the delta method) that contains a component of variance for
the number of marks observed and another for estimation of
p (Taylor et al., 2002). For cases where we stratified the
population into k strata (e.g., family class), the total estimated
population was the sum of stratum estimates and the variance
estimator was extended to include covariances between estimated capture probabilities in the k strata (see Taylor et al.,
2002). We used a similar estimator to construct a variance
estimate for the average population size over several years.
Our estimates of total survival (S) derived from capturerecapture data, included losses from harvest. We were
interested in estimating natural survival (Sn) to investigate
potential impacts of alternative harvest strategies. Unfortunately, total harvest of Baffin Bay bears from Greenland
was uncertain, so we could not use the approach of Taylor
et al. (2002) to estimate natural survival. Instead, we used
the following estimator:
Variances and covariances for p, S, and r were computed
by Program MARK after inflation by the overdispersion
coefficient.
Sn = S + (1 – S)r
where P = NR/NT. The threshold proportion (T) for a
specified recovery time (Y) was calculated as:
This estimator assumes that all harvested, marked bears
were recorded. Thus it may underestimate natural survival
if harvest of marked bears is underreported. But its advantage, compared to the estimator of Taylor et al. (2002), is
the constraint that Sn cannot exceed 1.0. Here, we also used
a Taylor series approximation for the variance of natural
survival:
Var(S n) = Var(S)(1 – r) + Var(r)(1 – S)2 + 2Cov(S,r)(1 – S)(1 – r)
Population Growth Rate
The geometric mean, zero-harvest population growth
rate (λN) and harvested population growth rate (λH) at
stable age distribution were calculated according to Taylor
et al. (1987b, 2001b). We ran 1200 Monte Carlo simulations
to estimate the geometric mean of λ using the life table–
based Visual Basic program ‘RISKMAN’ (Taylor et al.,
2001b; McLoughlin et al., 2003; McLoughlin and Messier,
2004) presented in the section on Harvest Risk Analysis.
We described variability about λ by presenting the
bootstrapped standard deviation (SD) of λ (Manly, 1997).
Harvest Risk Analysis
The typical metric for population viability analysis
(PVA) is mean likelihood of extinction or quasi-extinction
(i.e., effective extinction) over a set period of time (White,
2000). However, most managers of harvested populations
are not concerned about extinction, but rather about the
risk that overharvesting will reduce population numbers.
The management goal is typically to maximize harvest
opportunities for current hunters without placing the population and the interests of future generations at risk. We
used stochastic methods of PVA (e.g., Monte Carlo
simulations) to explore harvest management options for
Baffin Bay polar bears by estimating the risk of depletion
for a number of harvest scenarios. For this analysis, we
considered a population “depleted” if its size at the end of
the trajectory was below a threshold proportion (T) of the
initial population size that would not allow recovery within
a specified number of years if the harvest were to be
discontinued through a moratorium. We computed the
expected years to recovery (Y) from a reduced population
size (NR) to a target size (N T), in this case, initial population
size, using the unharvested population growth rate (λN):
Y=
−1n( P)
1n(λ N )
T = e-Y1n(λN)
A population reduced to P ≥ T could be expected to
recover to its original size within Y years if the harvest
were stopped (i.e., by a harvest moratorium). A population
reduced to P < T would not be expected to recover to its
original size within Y years even if the harvest were
stopped, which we defined as an unacceptable outcome.
This model conservatively assumes that at discontinuation
208 • M.K. TAYLOR et al.
of harvest, the natural (unharvested) rate of increase would
not be higher than that observed at the start of simulation.
Increasing harvests could, however, lead to decreased negative effects of density-dependence (i.e., a higher unharvested
rate of increase), for example, through increased reproduction or changes to age and sex distribution. We felt that
harvest risk levels using recovery times, rather than fractions
of the initial (original) population remaining, would make
consequences of overharvest more intuitive to managers and
the hunting communities they serve.
RISKMAN (Taylor et al., 2001b) was used to estimate
the proportion of acceptable outcomes (i.e., trajectories
that did not reduce the population below T) for a range of
recovery times from 1 to 25 years. RISKMAN differs from
other simulation models in providing the option to model
the specific population dynamics of species with multiyear
reproductive schedules, such as bears, cetaceans, elephants,
phocids, and primates (Taylor et al., 1987b), although
RISKMAN can also model simple annual reproduction.
Here we modeled age-specific recruitment rate as a function of three components: the availability (A) of a female
to mate at age x – 1 and produce offspring in year x, the
litter production rate of reproducing females (B), and the
litter size of newborns (L). Our definition of age-specific
recruitment rate (m x) was:
m x = Ax – 1 • Bx – 1• L x
RISKMAN allows the population to be structured into
any number of age classes with the following sex and
family-status categories: males; females with no young;
females with 1, 2, or 3 newborns; females with 1, 2, or 3
yearlings; and females with 1, 2, or 3 two-year-olds. The
model incorporates a minimum and maximum age of
reproduction. In addition to age- and sex-specific reproduction, age- and sex-specific survival, including maximum age of life (w), is a required input to the model, used
to compute age-specific survivorship (l x). We calculated
the population growth rate from the net reproductive rate
w
( R0 = ∑ lx mx ),
x=0
(the average number of offspring produced per individual
per lifetime), and generation length (Caughley, 1977).
RISKMAN incorporates stochasticity into its population model at several levels. To incorporate uncertainty in
initial population size, simulations can be generated using
a random initial population size (N T) drawn from a normal
distribution, with mean and standard error (SE) provided
by the user. For each year of simulation, RISKMAN
obtains a random normal deviate for each survival and
recruitment rate based on user-specified means and SEs
for particular sex and age strata. Individuals in the model
are then exposed in a series of Bernoulli trials to the
probabilities described by annual random deviates. This
process can incorporate annual variability and sampling
error, but also the uncertainty associated with applying the
random mean to individual trials, which result in either
success or failure (e.g., survival or death; a litter or failure
to produce a litter). Stochasticity in litter size and sex ratio
can also be incorporated into the model. RISKMAN uses
Monte Carlo techniques to generate a distribution of results (Manly, 1997), and then uses this distribution to
estimate the variance of summary parameters, such as
population size at a future time, population growth rate,
and the proportion of runs that result in a population
decline set at a predetermined level by the user. We
adopted the latter to evaluate threshold proportions (T) for
specified recovery times (Y). RISKMAN is available upon
request from the senior author.
Our scenarios were specific for a monitoring interval of
15 years beginning in 1998. We chose 15 years as a
simulation interval because we estimated that about 10%
of the Baffin Bay population would still be marked after 15
years. Adult survival rates are underestimated when there
are no observations of old marks in the population (i.e., no
observations of long life; M.K. Taylor, unpubl. data), so
we decided that our inventory monitoring cycle should not
exceed 15 years.
The proportion of acceptable outcomes was examined
for a range of annual harvest levels (40 – 180 bears/yr) that
were within the range of harvests likely to be experienced
by Baffin Bay polar bears (the historical annual kill for
Baffin Bay from 1970 to 1999 was estimated to average 88
bears per year; see Results). The frequency of occurrence
of acceptable outcomes (based upon the threshold value,
T) was monitored and reported as the cumulative proportion of successful runs over time. It was possible for a
simulated population to decline to an unacceptable number
and then increase back to an acceptable number before 15
years had elapsed. The cumulative frequency of acceptable values ranged between 0.0 and 1.0 and depended on
the starting population number, the survival and recruitment rates, the harvest sex and age composition, the
harvest number, and the uncertainty of these parameters
(all obtained from the previous standing age and markrecapture analyses). The unharvested population growth
rate determined the minimum time required for recovery
from any given reduction in numbers, and we stated this as
the proportion of original numbers required for recovery
times of 1, 5, 10, 15, 20, and 25 years in the absence of
further harvest.
For each year of the simulation, RISKMAN estimated
the mean and SE associated with total population size,
number of males, number of females, proportion of females in the harvest, population growth rate, and other
summary parameters. For our risk analysis, we identified
the initial age distribution by determining the stable age
distribution, using total mortality values. The harvest
selectivity-vulnerability array was identified by comparing the total mortality stable age distribution with the sex–
age–family status distribution of the historical harvest
(Taylor et al., 2002). The harvest sex–age–family status
POLAR BEAR DEMOGRAPHY AND POPULATION VIABILITY • 209
TABLE 1. Initial captures of unmarked polar bears classified by age and sex for 1974 – 92, 1993 – 95, and 1997. In parentheses, we present
numbers of bears recaptured at least once in the years 1993 – 95 and 1997, followed by the number of bears recovered in the harvest during
the harvest years 1975 – 99. Bears initially caught in 1997 had no chance of recapture (indicated by an asterisk), although they could have
been harvested in 1997 and 1998.
Sex
Female
Year
Cub
1 – 4 yr old
5 – 20 yr
1974 – 92
1993
1994
1995
1997
52 (2, 11)
15 (2, 3)
26 (6, 1)
16 (0, 4)
22 (*, 0)
131 (10, 19)
73 (10, 26)
22 (2, 5)
25 (5, 2)
34 (5, 3)
22 (*, 0)
176 (22, 36)
126 (12, 31)
30 (4, 2)
42 (7, 2)
37 (5, 1)
38 (*, 0)
273 (31, 36)
3 (1, 1)
1 (0, 0)
1 (0, 0)
2 (0, 0)
0 (*, 0)
7 (1, 1)
254 (28, 69)
68 (8, 10)
94 (18, 5)
89 (10, 8)
82 (*, 0)
587 (64, 92)
1974 – 92
1993
1994
1995
1997
52 (2, 16)
12 (0, 0)
16 (4, 0)
19 (0, 4)
19 (*, 0)
118 (6, 20)
67 (8, 33)
21 (11, 5)
25 (9, 4)
43 (6, 3)
32 (*, 5)
188 (34, 50)
81 (7, 26)
41 (18, 7)
52 (14, 6)
51 (6, 7)
86 (*, 3)
311 (45, 49)
4 (0, 0)
3 (1, 0)
7 (0, 0)
3 (1, 1)
0 (*, 0)
17 (2, 1)
204 (17, 75)
77 (30, 12)
100 (27, 10)
116 (13, 15)
137 (*, 8)
634 (87, 120)
Total
Male
Total
20 + yr
Total
TABLE 2. Estimates of the mean and standard error of reproductive
parameters of polar bears inhabiting Baffin Bay, 1994 – 97.
during 1974 – 98. These were largely legal kills made for
sport, for subsistence, or in the defense of life or property.
Parameter
Mean
SE
Reproduction
Cub litter size
Yearling litter size
Litter production rates for age 4
Litter production rates for age 5
Litter production rates for age 6+
Percentage of newborns that are male
Reproductive interval (years)
Reproductive rate (litters/year)
Proportion of females with only one litter
Proportion of females with no litters
1.59
1.63
0.10
0.88
1.00
0.49
2.55
0.39
0.12
0.05
0.07
0.07
0.12
0.40
0.17
0.03
0.14
0.02
0.07
0.03
distribution was stratified by sex, age (cubs and yearlings,
ages 2 – 4 years, ages 5 – 20 years, and ages 21 – 30 years),
and family status (alone, with cubs and yearlings, with
two-year-olds). Harvest simulations were conducted using the natural survival rate estimates (mark-recapture
analysis) and recruitment data (standing age analysis)
from this study; the harvest age and sex distribution for
Baffin Bay polar bears detailed in the 2001 Canadian
Federal and Provincial Polar Bear Technical Committee
(PBTC) status report (PBTC, 2001); and recovery rates of
marked animals. For starting population size (and SE), we
used the 1994 – 97 population estimate obtained from markrecapture analysis.
RESULTS
Captures, Recaptures, and Recoveries
Of 587 female and 634 male bears marked between
1974 and 1997, 64 females and 87 males were recaptured
at least once (Table 1). Of the 172 recaptures during 1993 –
97, 49 were recaptures of radio-collared female bears or
their accompanying young. In addition, 212 marked bears
(92 females and 120 males) were recovered as dead bears
Summary reproductive parameters for the Baffin Bay
polar bear population based on analysis of the standing age
distribution for captures (1994 – 97) are presented in Table 2.
Mark-Recapture Analysis
Of the 935 total bear captures and recaptures from 1993
to 1997, 145 involved cubs; hence, we used a value of ĉ =
1.18 for overdispersion in our mark-recapture analysis
(i.e., one plus the ratio of cub captures to non-cub captures). The lowest QAICc model (S/rage + sex + season, pradio +
family) included additive effects of age, sex, and marking
season on survival and recovery. Capture probability was
constant across time (1993 – 97) but varied across family
classes, and our use of radio locations made capture almost
certain for radio-collared females (Table 3). The next three
models with the lowest ∆QAICc values (Table 3) were
sufficiently close that model averaging was indicated
(Burnham and Anderson, 2002), and we included these
models with the primary model in a weighted average to
present survival and abundance estimates.
Estimates of model coefficients provide for interpretation of covariate effects (Table 4). The probability of
autumn recapture was lower for females and yearling cubs
than for adult males and subadults, except for radiocollared females and their young. The estimated recovery
rate was highest for subadults and males, the two age-sex
strata most likely to be harvested. Cubs and older bears
(21+) were least likely to be hunted. Survival rate was
lower for bears caught in spring than for bears caught in
autumn, likely as a result of higher harvest rates for springcaught bears.
We present model-averaged survival estimates only for
bears marked in autumn (Table 5). This group is more
210 • M.K. TAYLOR et al.
TABLE 3. ∆QAICc values for models fitted to capture-recapture and harvest recovery data for polar bears in Baffin Bay (1974–97).
Covariates of the survival/recovery model included age and sex of bears and season of recovery (autumn or spring). Covariates of the capture
probability model included age and sex of captured animals, year of capture or recovery, and whether or not the animal was wearing a radio
transmitter (radio). The number of estimated parameters for each sub-model is shown in parentheses. Values in bold type represent models
used in model averaging.
Survival/Recovery Model (S/r)
Capture Probability Model (p)
radio + year (4)
radio + family (3)
radio (1)
age + season (10)
age + sex + season (12)
age × sex + season (16)
9.75
3.54
8.01
12.69
6.72
11.01
TABLE 4. Estimates and standard errors (SE) of logit parameters
for the model with minimum QAICc. Real estimates can be computed
with the inverse logit: exp(x)/(1+exp(x)). Each effect is additive in
relation to the intercept. For example, x = 0.448 + 1.840 – 0.307 for
survival of one- to four-year-old males. The SE for the radio effect
was not reliable because the parameter was at a boundary (p = 1.0).
An estimate divided by its SE can be treated as an approximate
t-value in a significance test that the estimate is non-zero (Burnham
and Anderson, 2002).
Sub-model
Survival
Capture
Recovery
Logit Coefficient
Intercept (Female cub)
1–4
5 – 20
21+
Male
Spring
Intercept (Male adult)
Radio
Adult female + yr
1–4
Intercept (Female cub)
1–4
5 – 20
21+
Male
Spring
Estimate
0.448
1.840
2.339
1.868
-0.307
-0.718
-1.560
21.760
-0.648
-0.429
-2.973
2.479
1.723
0.358
0.474
0.677
SE
0.365
0.477
0.334
0.492
0.189
0.293
0.152
N/A
0.222
0.245
0.443
0.669
0.388
0.578
0.228
0.341
7.26
0
4.27
radio + family + year (6)
10.72
3.72
7.93
TABLE 5. Mark-recapture estimates of survival for the Baffin Bay
polar bear population, 1994 – 97 (autumn-marked bears only). The
upper rows present the mean and SE of natural survival (no harvest)
for the various sex and age categories, and the lower rows, those of
actual survival (including harvest mortality).
Female Survival Rate
Cub 1 – 4 5 – 20 21+
Male Survival Rate
Cub 1 – 4 5 – 20 21+
Mean
SE
0.620 0.938 0.953 0.919
0.095 0.042 0.020 0.046
0.570 0.938 0.947 0.887
0.094 0.045 0.022 0.060
Mean
SE
0.600 0.901 0.940 0.913
0.096 0.045 0.021 0.047
0.538 0.879 0.923 0.874
0.094 0.049 0.024 0.062
Estimate/SE
1.2
3.9
7.0
3.8
-1.6
-2.4
-10.2
N/A
-2.9
-1.8
-6.7
3.7
4.4
0.6
2.1
2.0
representative of the entire population because it includes
those bears that did not frequent shorefast ice in spring.
The clearest age and sex difference was between cub and
non-cub survival. While the model included possible senescence (survival estimates were lower for bears aged 21
years or more than for other non-cub classes), the survival
rate of bears in the oldest age class was within 1 SE of the
rate for bears aged 5 – 20 years (Table 5). Similarly, survival rates were not substantially different for males and
females (Tables 4 and 5), but harvest recovery rates were
substantially higher for males than for females (Table 4).
The estimated natural survival rates for males and females
5 – 20 years old also differed by less than 1 SE (Table 4).
The model-averaged, mark-recapture estimate of mean
abundance for years 1994 – 97 was 2074 bears (SE = 266),
which included 1017 females (SE = 192) and 1057 males
(SE = 124). Females of age 5 years or more averaged 524
(SE = 108) individuals. The total number of cubs in the
population was estimated to be 387 (SE = 82), giving an
average natality of 0.74 cubs per adult female.
Population Growth Rate
Using mark-recapture survival rates and reproduction
data, our bootstrapped estimate of the geometric mean
unharvested, stable age population growth rate (λN) was
1.055 (SD = 0.011). Assuming the 1970 – 99 average
annual Canadian harvest of Baffin Bay polar bears (66
bears/yr), and the probable recovery of marked polar bears
by Greenland hunters (15 – 25 bears/yr: we used 22 bears/
yr as an average), we estimated the geometric mean harvested, stable age population growth rate (λH) to be 1.019
(SD = 0.015).
Harvest Risk Analysis
The proportion of outcomes for the range of simulated
harvest levels (40 – 180 bears/yr) and range of recovery
times (1 – 25 yrs) for the Baffin Bay population are provided as a smoothed contour map (Fig. 2). Higher harvest
rates constituted increased risk (fewer acceptable outcomes) to the population. If managers and stakeholders are
willing to accept long periods (e.g., decades) of harvest
moratorium should a harvest threshold be exceeded, the
proportion of acceptable outcomes increases. To take an
example using the estimated harvest of polar bears in
Baffin Bay (88 bears/yr), the probability that the population would either increase or decline no more than could be
recovered in five years was 0.95. That is, 95% of the time,
the population was not expected to decline below that
proportion of the starting population required, given our
estimate of natural (i.e., unharvested) population growth
POLAR BEAR DEMOGRAPHY AND POPULATION VIABILITY • 211
FIG. 2. The proportion of acceptable simulation outcomes (contour values) for
a range of recovery times (1 – 25 years) and a range of annual harvest values
(40 – 180). “Acceptable” in this context means that in a given simulation run,
the population did not decline to a level that would require more than x years
to recover.
rate (λN = 1.055), to recover to the original population size
(2074 bears) within five years. The likelihood that the
population might require more than five years to recover
was 0.05 (i.e., 1 minus 0.95). If the harvest was raised to
120 bears per year, the chances that the population would
either increase or decline no more than could be recovered
in five years was reduced to approximately 0.81 (Fig. 2).
DISCUSSION
Observed natural (i.e., unharvested) survival rates and
reproductive parameters for polar bears in Baffin Bay
were generally higher than those obtained for adjacent
populations of polar bears, suggesting that the Baffin Bay
population is below carrying capacity and relatively vigorous. The unharvested finite rate of increase (λN = 1.055)—
the summary parameter combining data on natural survival
and reproduction—was higher than that estimated (using
the same methods presented here) for polar bears in Kane
Basin to the northeast (λN = 1.009) and in Lancaster Sound
(λN = 1.030) and Norwegian Bay (λN = 1.011) to the
northwest (M.K. Taylor, unpubl. data). Only the
unharvested population growth rate for polar bears in
Viscount Melville Sound (λN = 1.059; Taylor et al., 2002),
a population reduced substantially below carrying capacity by years of overharvest, was higher than the unharvested
finite rate of increase of Baffin Bay polar bears. It is
difficult to compare parameters presented in this study
with those of other polar bear populations because of
different methods and different terminology (Stirling et
al., 1980, 1999; Ramsay and Stirling, 1988; Derocher and
Stirling, 1992, 1995; Amstrup, 1995). Further, few researchers have been able to estimate the rates of natural
survival necessary for useful ecological comparisons (e.g.,
life history comparisons). However, our estimates of reproductive parameters (e.g., litter size and timing of first
reproduction) fall within the general range of published
values. Taylor et al. (2002) present a comprehensive comparison of vital rates among studied polar bear populations.
Taylor et al. (2002) suggested that previous estimates of
vital rates and abundance for polar bears may have been
compromised by the failure to incorporate heterogeneity
into capture probabilities. Our earlier attempt to quantify
the abundance of polar bears in Baffin Bay likely failed in
this respect. Using mark-recapture data from 1980 to
1985, abundance was initially estimated (M.K. Taylor,
unpubl. data) using a variety of open Jolly/Seber methods
(Jolly, 1965; Seber, 1965; Pollock et al., 1990). At that
time, we had yet to define population bounds for Baffin
Bay polar bears (Taylor et al., 2001a) and relied only on
spring-caught (i.e., on-ice) bears. Our earlier estimates of
abundance for bears in Baffin Bay were markedly lower
than those presented here. The highest estimate for the
period 1980 – 85 was 937 bears (SE = 265). This estimate
of population size indicates that the quotas established
during the late 1980s and 1990s were likely made in error.
The quotas were developed against the advice of Inuit
hunters, who felt that the population on which quotas were
based was much larger than estimated.
Following the recommendation of local hunters, we
conducted our most recent mark-recapture inventory in
autumn (1994 – 97), when bears were on shore at Baffin
Island. We were able to satisfy assumptions of a more
flexible analysis model, and our estimate of abundance
was considerably higher (i.e., 2074 bears, SE = 266). By
capturing every bear we encountered during autumn, we
eliminated the limitation of previous sampling efforts
conducted only in spring months, when most of the population was distributed on active pack ice and unavailable to
capture teams. Our best information suggests that discrepancies in estimated abundance between the 1980 – 85 and
1994 – 97 studies were not the result of rapid population
growth between census periods. Traditional ecological
knowledge suggested stability in the population from 1985
to 1994 (M.K. Taylor, unpubl. data); data presented here
suggest that from 1994 to 1997 the population was stable
or slightly increasing (λH = 1.019).
As expected (Taylor et al., 2002), all of the top models
(Table 3) included the radio-collar covariate (or being
associated with a radio-collared bear) as an important
predictor of capture probability. Female polar bears
equipped with radio-collars and their cubs were more
likely to be recaptured than other bears. Some bears were
located and recaptured using radio signals; however, even
after radio-collars were removed, these females and their
cubs continued to have a higher probability of recapture
than other bears. Behavioural preference of radio-collared
bears for geographic areas where individuals are more
212 • M.K. TAYLOR et al.
easily observed (and thus captured), e.g., their affinity to
landmarks and the sea coast, is one possible explanation of
this result.
Analyses of other populations had led us to anticipate
sex and age stratification for survival probability of cubs
(age 0), subadults (age 1 – 4), adults (age 5 – 19), and older
adults (age 21+); however, the only clear differences were
lower survival for cubs and for bears caught in spring, and
higher harvest recovery rates for males, prime-aged bears,
and bears caught in spring (Tables 4 and 5). Lower survival
rates and higher harvest recovery rates for bears marked in
spring were consistent with previous findings that bears
captured in spring represent a portion of the Baffin Bay
population that is more susceptible to harvest (M.K. Taylor,
unpubl. data). Our failure to demonstrate a decline in
survival of older bears was likely due to insufficient
sample size.
The population number (N > 2000) imposes an obvious
constraint on the capture sample size, and small recapture
samples relative to actual population size may limit analysis to models with fewer parameters. Data presented here
were the result of a capture-recapture program spanning
three decades, with a large effort exerted during the autumn captures of 1994–97. Even with the dedicated autumn onshore program, however, recapture probabilities
ranged from 0.11 to 0.17 for bears without radio-collars.
Faced with sample size limitations, it is essential to incorporate all sources of data into an analysis. By including
harvest recoveries, we were able to double the number of
bears “recaptured” after their initial release, which enabled us to derive more precise estimates of survival and
abundance for risk assessment and management advice.
However, we have yet to use our data to their fullest extent
because we did not incorporate recoveries of unmarked
bears or first captures of unmarked bears. At present, there
are no models or software programs that provide a joint
analysis of recovery-captures/recaptures using the first
encounter (capture/harvest). Development of a more general model within a cohort framework is currently being
investigated.
A stable harvested population growth rate, together
with a relatively high unharvested population growth rate,
suggests polar bears in Baffin Bay are being sustainably
harvested below food carrying capacity; however, the
existing harvest still presents some risk to the population.
Our fieldwork was carried out in collaboration with local
hunters, and it was immediately apparent that hunters were
very adept at locating and harvesting polar bears. Although the study area was large and density of polar bears
relatively low, local hunters maintained high success rates
in all years. This hunt was legal and within the quota
limitations set by the Governments of Greenland and
Nunavut, Canada. Although we were able to document
that the current harvest rate (60 – 80 bears/yr) is likely
sustainable, our information also suggests that there is no
safe maximum sustainable harvest rate for this population.
Our risk assessment is simply an expression of uncertainty
in the demographic parameters on which it is based. The
RISKMAN simulations indicated that even though the
estimated population growth rate in the absence of harvest
was high, the population is at some risk even if the annual
harvest policy is low (Fig. 2).
We believe our harvesting simulations are much more
realistic than a deterministic calculation of maximum
sustained yield that does not consider the uncertainty of
underlying information. Both managers and stakeholders
must recognize that scientific information rarely provides
exact and absolutely correct harvest rate or quota values.
Researchers have a responsibility to quantify the uncertainty of their measurements, and also the uncertainty of
their management recommendations. Reporting scientific
results in this manner is not only more honest; it also
identifies the distribution of solutions where local and
traditional knowledge may be used to make final
determinations.
A potential problem with our harvest analysis is that our
estimated take of bears by Greenland hunters (18–25
bears/year), which was derived from recovery probabilities of marked bears, is lower than the number of bears
reported as harvested in Greenland. In some years, systematic polar bear harvest surveys were undertaken by the
Greenland Home Rule Government (E.W. Born, pers.
obs.). Although the sex and age composition of the Greenland harvest was not recorded during these surveys, the
estimated harvest was approximately 72 bears per year
(E.W. Born, unpubl. data). The large discrepancy between
this estimate of Greenland’s hunt and the estimate from the
harvest of marked bears raises some concerns. It is possible our sampling protocol was biased in undersampling
bears that, for whatever reason, were more susceptible to
Greenland harvest, or that Greenland overreported kills in
their surveys (e.g., if two or more hunters who shared a kill
reported it independently when surveyed). Our own
simulations show that if the Greenland kill were indeed 72
bears/year (i.e., combined Greenland/Nunavut take is 138
bears/year), the population would be declining (λH = 0.974,
SD = 0.041), and a lengthy moratorium would be required
to allow the population to recover from this decline
(Fig. 2). However, Baffin Bay hunters have been emphatic
that the population is increasing, which is consistent with
our simulations. We suggest that confidence in polar bear
conservation decisions would be enhanced by more systematic collection of Greenland’s harvest data.
Until the Greenland kill can be verified and confirmed,
our management recommendations must remain qualified.
Management decisions must also consider the long-term
goals for the population and the resources available for
monitoring. Shorter intervals between population inventories could allow more aggressive harvest policies with the
same level of risk, but these inventories are expensive,
invasive, and time-consuming. Our information suggests
that no matter what harvest level is chosen, stakeholders
should understand that estimates and management recommendations are based on uncertain data, and that if the
POLAR BEAR DEMOGRAPHY AND POPULATION VIABILITY • 213
population should decline, long-term harvest moratoriums
may be required to regain the current number of bears.
ACKNOWLEDGEMENTS
Financial support for this work was provided by the Greenland
Home Rule Government, the Government of the Northwest
Territories, the Government of Nunavut, Parks Canada, local
Hunters’ and Trappers’ Organizations, the Nunavut Wildlife
Management Board, the Polar Continental Shelf Project, the
University of Saskatchewan, and the U.S. National Marine Mammal
Laboratory. Hunters of Pond Inlet, Clyde River, Qikiqtarjuak, and
Savissivik contributed their local and traditional knowledge of
polar bears and participated in the work and harvest program. We
particularly thank Sam Palituq, David Kooneeliusie, and Jako
Allooloo for their assistance in the field.
REFERENCES
AMSTRUP, S. 1995. Movements, distribution, and population
dynamics of polar bears in the Beaufort Sea. Doctoral dissertation,
University of Alaska, Fairbanks.
ARVESON, J.N. 1969. Jackknifing U-statistics. Annals of
Mathematics and Statistics 40:2076 – 2100.
BURNHAM, K.P. 1993. A theory for combined analysis of ring
recovery and recapture data. In: Lebreton, J.D., and North, P.M.,
eds. Marked individuals in the study of bird populations. Basel,
Switzerland: Birkhäuser. 199 – 213.
BURNHAM, K.P., and ANDERSON, D.R. 2002. Model selection
and inference: A practical information-theoretic approach. New
York: Springer-Verlag.
CALVERT, W., and RAMSAY, M.A. 1998. Evaluation of age
determination of polar bears by counts of cementum growth
layer groups. Ursus 10:449 – 453.
CAUGHLEY, G. 1977. Analysis of vertebrate populations. New
York: John Wiley & Sons.
DEROCHER, A.E., and STIRLING, I. 1992. The population
dynamics of polar bears in western Hudson Bay. In: McCullough,
D.R., and Barret, R.H., eds. Wildlife 2001: Populations. London:
Elsevier Applied Science. 1150 – 1159.
———. 1995. Temporal variation in reproduction and body mass
of polar bears in western Hudson Bay. Canadian Journal of
Zoology 73:657 – 1665.
FERGUSON, S.H., TAYLOR, M.K., and MESSIER, F. 1997.
Space-use of polar bears in and around Auyuittuq National Park,
Northwest Territories, during the ice-free period. Canadian
Journal of Zoology 75:1585 – 1594.
FERGUSON, S.H., TAYLOR, M.K., ROSING-ASVID, A., BORN,
E.W., and MESSIER, F. 2000. Relationships between denning
of polar bears and conditions of sea ice. Journal of Mammalogy
81:1118 – 1127.
FURNELL, D.J., and SCHWEINSBURG, R.E. 1984. Population
dynamics of central Arctic polar bears. Journal of Wildlife
Management 48:722 – 728.
JOLLY, G.M. 1965. Explicit estimates from capture-recapture data
with both death and immigration stochastic model. Biometrika
52:225 – 247.
LEBRETON, J.D., BURNHAM, K.P., CLOBERT, J., and
ANDERSON, D.R. 1992. Modelling survival and testing
biological hypotheses using marked animals: A unified approach
with case studies. Ecological Monographs 62:67 – 118.
MANLY, B.F.J. 1997. Randomization, bootstrap and Monte Carlo
methods in biology. London: Chapman & Hall.
McCULLAGH, P., and NELDER, J.A. 1991. Generalized linear
models. London: Chapman & Hall.
McDONALD, T.L., and AMSTRUP, S.C. 2001. Estimation of
population size using open mark-recapture models. Journal of
Agricultural, Biological, and Environmental Statistics 6:
206 – 220.
McLOUGHLIN, P.D., and MESSIER, F. 2004. Relative
contributions of sampling error in initial population size and
vital rates to outcomes of population viability analysis.
Conservation Biology 18:1665 – 1669.
McLOUGHLIN, P.D., TAYLOR, M.K., CLUFF, H.D., GAU, R.J.,
MULDERS, R., CASE, R.L., and MESSIER, F. 2003. Population
viability of barren-ground grizzly bears in Nunavut and the
Northwest Territories. Arctic 56(2):185 – 190.
PAETKAU, D., AMSTRUP, S.C., BORN, E.W., CALVERT, W.,
DEROCHER, A.E., GARNER, G.W., MESSIER, F., STIRLING,
I., TAYLOR, M.K., WIIG, Ø., and STROBECK, C. 1999.
Genetic structure of the world’s polar bear populations. Molecular
Ecology 8:1571 – 1584.
PBTC (POLAR BEAR TECHNICAL COMMITTEE). 2001.
Minutes of 2001 meeting of the Federal and Provincial Polar
Bear Technical Committee, 4 – 5 February 2001. Available from
the Canadian Wildlife Service, 5320 – 122 Street, Edmonton,
Alberta T6H 3S5.
POLLOCK, K.H., NICHOLS, J.D., BROWNIE, C., and HINES,
J.E. 1990. Statistical inference for capture-recapture experiments.
Wildlife Monographs 107. 97 p.
RAMSAY, M.A., and STIRLING, I. 1988. Reproductive biology
and ecology of female polar bears (Ursus maritimus). Journal of
Zoology (London) 214:601 – 634.
SEBER, G.A.F. 1965. A note on the multiple recapture census.
Biometrika 52:249–259.
_____. 1982. The estimation of animal abundance and related
parameters. New York: Macmillan.
STIRLING, I., CALVERT, W., and ANDRIASHEK, D. 1980.
Population ecology studies of the polar bear in the area of
southeastern Baffin Island. Edmonton: Canadian Wildlife Service
Occasional Paper No. 33.
STIRLING, I., SPENCER, C., and ANDRIASHEK, D. 1989.
Immobilization of polar bears (Ursus maritimus) with Telazol in
the Canadian Arctic. Journal of Wildlife Diseases 25:159 – 168.
STIRLING, I., LUNN, N.J., and IACOZZA, J. 1999. Long-term
trends in the population ecology of polar bears in western
Hudson Bay in relation to climatic change. Arctic 52(3):
294 – 306.
TAYLOR, M., and LEE, J. 1995. Distribution and abundance of
Canadian polar bear populations: A management perspective.
Arctic 48(2):147 – 154.
214 • M.K. TAYLOR et al.
TAYLOR, M.K., BUNNELL, F., DeMASTER, D.,
SCHWEINSBURG, R., and SMITH, J. 1987a. ANURSUS: A
population analysis system for polar bears (Ursus maritimus).
International Conference on Bear Research and Management
7:117 – 125.
TAYLOR, M.K., CARLEY, J.S., and BUNNELL, F.L. 1987b.
Correct and incorrect use of recruitment rates for marine
mammals. Marine Mammal Science 3:171 – 178.
TAYLOR, M.K., KUC, M., and ABRAHAM, D. 2000. Vital Rates:
Population parameter analysis program for species with threeyear reproductive schedules. Iqaluit: Government of Nunavut.
TAYLOR, M.K., AKEEAGOK, S., ANDRIASHEK, D.,
BARBOUR, W., BORN, E.W., CALVERT, W., FERGUSON,
S., LAAKE, J., ROSING-ASVID, A., STIRLING, I., and
MESSIER, F. 2001a. Delineating Canadian and Greenland
polar bear (Ursus maritimus) populations by cluster analysis of
movements. Canadian Journal of Zoology 79:690 – 709.
TAYLOR, M.K., OBBARD, M., POND, B., KUC, M., and
ABRAHAM, D. 2001b. RISKMAN: Stochastic and deterministic
population modeling RISK MANagement decision tool for
harvested and unharvested populations. Iqaluit: Government of
Nunavut.
TAYLOR, M.K., LAAKE, J., CLUFF, H.D., RAMSAY, M., and
MESSIER, F. 2002. Managing the risk from hunting for the
Viscount-Melville Sound polar bear population. Ursus 13:
185 – 202.
WHITE, G.C. 2000. Population viability analysis: Data requirements
and essential analyses. In: Boitani, L., and Fullers, T.K., eds.
Research techniques in animal ecology: Controversies and
consequences. New York: Cambridge University Press.
288 – 331.
WHITE, G.C., BURNHAM, K.P., and ANDERSON, D.R. 2000.
Advanced features of program MARK. Fort Collins: Colorado
State University.